Multi-type random game dynamics: limits at discontinuities and cyclic limits

TL;DR

This paper analyzes complex population dynamics involving mixed rational and behavioral players using a differential inclusion approach, revealing new limit behaviors at discontinuities and conditions for cyclic outcomes.

Contribution

It introduces a stochastic-approximation framework for discontinuous dynamics, characterizes limits via ICT sets, and explores cyclic outcomes in strategic population games.

Findings

Identification of non-classical zeros at discontinuities

Conditions for cyclic outcomes in population dynamics

Impact of player types on long-term strategic behavior

Abstract

We consider (random) strategic interactions in a large population consisting of a variety of players. A rational player chooses actions that maximize certain utility functions, while a behavioral player chooses actions based on preferences such as avoid-the-crowd or follow-the-majority. We specifically study a turn-by-turn dynamic process in which players choose their actions sequentially and once; the utilities are realized either immediately or at the end of the game. In the literature, such dynamical systems are often analyzed using an appropriate approximating ordinary differential equation (ODE). However, the ODEs approximating the dynamics with pure actions are typically discontinuous. We adopt a differential inclusion (DI) based stochastic-approximation framework to derive the limiting analysis. The limits of the dynamics are characterized through the internally chain transitive (ICT) sets. We identify the presence of non-classical zeros as potential limits of the dynamics, a phenomenon not observed in classical settings involving continuous ODEs. These new limits arise precisely at the points of discontinuity of the dynamics. We further provide the conditions under which cyclic outcomes may occur at the limit. Finally, we study a queuing game with differential priority-based services and examine the impact of the proportions of avoid-the-crowd and two types of rational populations on the long-run outcomes of the strategic interactions. We identify potential point limits and establish the possibility of cyclic outcomes for certain parameter configurations.